应用粒计算的混合智能故障诊断技术研究

Hybrid Intelligent Diagnosis Technology Based on Granular Computing

针对现有的混合智能故障诊断模型缺乏通用方法和混合框架,未能实现不同智能诊断方法的实质性融合和优势互补的问题,提出并构建了一种基于粒计算的混合智能故障诊断模型.该模型的核心是在邻域粗糙集中求取不同的邻域值,对故障特征集进行分层粒化,在不同粒度下获得核属性集.利用核属性集在相应粒度下构建人工神经网络和支持向量机子分类器,通过评估矩阵算法对所有粒度下全部子分类器的诊断结果进行融合集成.模型应用结果表明,分类精度随着粒度层的增加而不断提高,集成后的分类精度高于不同粒度下的所有子分类器,从而体现了粒化分层的优势和不同智能诊断方法的优势互补,为混合智能诊断提供了一种新途径.

Aiming at lacking hybrid modes and common algorithms in existing hybrid intelligent diagnosis, a new model of hybrid intelligent fault diagnosis based on granular computing is pro- posed. In the model, the core features sets （CFS） are extracted in different granular levels by the reduction algorithm based on neighborhood rough set, then, CFS are chosen to train artificial neural network and support vector machines as sub-classifiers in corresponding levels. And the results of sub-classifiers in different granular levels are combined by criterion matrix algorithm as output of hybrid intelligent diagnosis. The model is applied to fault diagnosis of roller bearings in high-speed locomotive. The application results demonstrate that the classification accuracy is raised with the increasing grantilar levels, and the accuracy of hybrid results is higher than the one of any sub-classifier. The proposed model exhibits the effect of granulation and the advantages complementation among different intelligent methods to provide a new way for hybrid intelligent diagnosis.